Author: Uri Blass
Date: 13:26:01 09/07/01
Go up one level in this thread
On September 07, 2001 at 15:28:31, Christopher R. Dorr wrote: >On September 07, 2001 at 15:08:17, Uri Blass wrote: > >>On September 07, 2001 at 09:43:00, Christopher R. Dorr wrote: >> >>>I really do not see these conditions as a huge advantage for Kramnik. >>> >>>1. If Kramnik was going to play a serious match against a human GM, he would >>>study all the games by him/her in depth. >> >>seeing games is not the same as having the program. > > >OK. They also had the option of producing a body of 500-1000 tournament and >match games between DF and GMs. They had not this option Computers are almost not allowed to play against humans. That's what would have happened had Kramnkik >been playing a million dollar match against 8any other GM on earth*. They >didn't. They couldn't. You want *Kramnik* at a disadvantage? I think that seeing comp-comp games is enough. the computer does not see a lot of games of kramnik against machines so kramnik does not need to see a lot of games of Deep Fritz against humans. > >> >> Most human GMs would have hundreds or >>>thousands of games to study. Deep Fritz does NOT have these games yet. A >>>reasonable alternative would be to allow the GM to use the program before hand. >>>The CB team certainly has studied *all* of Kramnik's games closely. >> >>I am not sure about it and I doubt if it could help them. >>I suspect that working on improving the engine >>could be more productive for them then working on kramnik's games. >> >>Chessbase know that Deep Fritz will lose the match and it is totally >>unimportant for them to waste time about kramnik's games before >>the match. > >So the match is unfair because the CB teams decides not to use the information >they have at their disposal? > >> >> Why should >>>the computer have the advantage of knowing the GMs style and weaknesses when the >>>GM does not get to understand the computers? >> >> >>The computer can know only games of previous version of kramnik >>so it has nop advantage. >> > >Previous version? You think human GMs change evaluation parameters, styles, >openings, positional preferences at will? Hardly. Look at Karpov's early games. >Look at his present games. See a pattern? There isn't a Karpov 2.0 vs. a Karpov >6.3. The CB team knows what to expect with Kramnik. They know his style. They >know that unless his opening book alters, he will play X in response t Y because >in 85% of previous opportunities to do so, he did. > >>kramnik can also get games of previous versions of Fritz and >>I guess that the difference between Deep Fritz and the Fritz >>that plays against kramnik is smaller than the difference between >>the Kramnik who will play against Deep Fritz and the previous kramnik. >> > > >Highly doubtful. Was the Kasparov who played Karpov in 1985 much different from >the Kasparov of 1986? Not really. Similar style, similar evaluations. Kasparov who played against computers was clearly different than kasparov who played against humans and this is the point here. You don't >forget an entire lifetime's chess preperation and understanding simply because >you are playing an opponent you never have before. > >>> >>>2. Fritz can change it's book at will. If there *are* any 'predetermined' games, >>>they arise from the opening. Modify the book, elimninate the problem. >>> >>>3. And I doubt there could be predetermined games anyway. Think about the >>>variety of openings. Kramnik is going to *find* and *memorize* killer lines in >>>everything? As black, he would have to prepare for at least 12 major openings by >>>white, and literally hundreds of important subvariations (If he chooses to play >>>a Scheveningen Sicilian, for example, you need to worry about Keres attack with >>>h3, Keres Attack with Rg1, Classical, Bc4, King's Indian Attack......). And >> >>Why do you assume that he chooses sicilian? >>Choosing openings like 1.e4 h5 makes more sense if you want to win by >>memorizing games. > >And that would be the fault of who? The CB team who didn't plan for this and >write some trash openings into the opening book? I helped out on the opening >book for NOW several years ago before a Harvard Cup. Even then, I threw in some >trash lines (for it to respond to, not play), just in case an opponent wanted to >play 1.h4 just to throw it off. Giving a reply to 1.h4 is not enough because the opponent may prepare 20 games against 1.h4 a5,1.h4 a6... You need to build a bigger book if you are afraid of preperation. It was not important in the harvard cup because the players did not train 3 monthes against the program. I agree that it is possible to do it. > >> >>>remember, there are already very few *big* holes in Fritz's opening book. The >>>*small* ones Kramnik would find would likely be further down the tree, making it >>>even more unlikely that he could spring a particular trap. >> >>The problem is not holes in the opening book but holes in some >>positions when kramnik can push the machine to play. >>> >>> >>>Overall, I do not see Kramnik as getting a huge advantage. He didn't get the >>>source code. >> >>He can see the evaluation of Deep Fritz and learns from it >>about the source code. > >And the CB team can read his annotations and analyse what he likes to do. Do you >think the evaluation stuff for Deep Fritz is so unique that he couldn't do very >good prep using Fritz 5? Or Crafty? DF evals a position at +1.02 after 12 ply, >Fritz 5 eval'd the same position +0.99 after 8 ply. What did he learn from this? He can see also the main line and he can learn for example that Fritz overevaluate passed pawns in some kind of positions. >That DF searches deeper? I'm guessing he figured that out already. > > >> >> The program isn't under a 'truth serum'. He has access to it's >>>play, just as CB has access to all of Kramnik's games. As matches go, I see this >>>one as fairly equal in it's treatment of the two players. >> >>"just as?" > >Basically, yes. if the CB team didn't like the conditions (i.e. thought they >were unfair) they certainly had the chance to decline the match. It seems that they think that they can do a lot of money from losing the match so they do not care about it. > >> >>Kramnik learned from the games that he played and he is probably not >>going to do the same kind of mistakes again. >>Deep Fritz can learn nothing from the games that kramnik can see >>and not because of a weakness of the program but because of the >>fact that it cannot know about these games. >> >>Kramnik can learn more than seeing games of the machine. >>If seeing games of the machine was the point that it was enough >>to give kramnik a printout of Deep fritz games against other machines. >> >>Uri > >Kramnik didn't *need* to see games vs other boxes. he needed to see games vs. >human GMs. Deep Fritz does not need to see games of kramnik versus other humans. It needs to see games of Kramnik versus other boxes. DF didn't have those. A reasonable accommodation is to then let him >play with the program, and have his GMs play with it. > >It seems to me many of those complaining about the unfairness of the match are >simply setting it up so that if Kramnik won, they can say that it means nothing. >It *does* mean something. In the end, a chess game is a contest of a finite >length between two players. If DF plays better, it will likely win. if Kramnik >plays better, he will likely win. If Kramnik knows the machine's weaknesses and >can coax it into playing worse, then more power to him. That's what the game is >supposed to be about. If DF can figure out Kramnik's weaknesses and can do the >same, fantastic! Deep Fritz cannot do it because of the unfair conditions. Suppose Deep fritz can do it. Kermnik is going to discover it after having the program and learn to avoid the weaknesses. If kramnik can take advantage of the program's weaknesses the program has no way to learn it from games against kramnik before the match in order to fix the problem. Uri
This page took 0 seconds to execute
Last modified: Thu, 15 Apr 21 08:11:13 -0700
Current Computer Chess Club Forums at Talkchess. This site by Sean Mintz.